RM-83 4.5.2(5,6,7) and RM-95 4.5.2(24) ... Equality ...
[rest unintelligible] ... Compound objects ...
[rest unintelligible] ... Tagged types ...
[rest unintelligible] ... Generics ...
[rest unintelligible] ...
A large and dark room. In the foreground a terminal, at the rear wall a huge screen showing everything that happens at the terminal. At first the screen is off, the room in twilight. Below the stage, but visible, Lady Ada in a coffin. Around the stage on a gallery, nearly invisible in the dark (also when the stage is lit), the chorus of Ada gurus.
Seated on a swan, Tucker Taft hovers down from the ceiling, murmuring some unintelligible quotes from the Ada Reference Manual:
RM-83 4.5.2(5,6,7) and RM-95 4.5.2(24) ... Equality ... |
|
The screen brightens, putting the room into an eerie light, and a lake appears on it. Another swan enters the picture, carrying John Goodenough. Tucker's swan penetrates the screen and, with a splash, lands in the lake beside John Goodenough.
Tucker Taft startles and shouts:
Heureka! Liberty, Equality.
Brother, we need equality for all!
John Goodenough declaims:
It is already good enough, |
 |
Lady Ada suffers convulsive jerks. Tucker Taft soars off, the swan with John Goodenough swims away, and the screen grows dim, leaving the room again in twilight, while the chorus chants:
Fair is foul and foul is fair
Hover through the filthy air.
Spotlights on, the room becomes as light as day. Enter Knopp (a programmer). He sits down in front of the terminal and starts writing; the screen at the wall displays:
type Container is record
Length : Natural;
Content: String (Some_Range);
end record
function "=" (Left, Right: Container) return Boolean is
begin
return Left .Content (1 .. Left .Length) =
Right.Content (1 .. Right.Length);
end "=";
declare
T: constant String := "Lady Ada";
CON_1, CON_2: Container;
B_Con: Boolean;
begin
CON_1.Length := T'Length; CON_1.Content (1 .. T'Length) := T;
CON_2.Length := T'Length; CON_2.Content (1 .. T'Length) := T;
B_Con := CON_1 = CON_2;
end;Knopp starts a program, the result is shown on the screen:
CON_1 = CON_2 returns True! (qed)
Chorus chants:
Swiftly Knopp evades this place
And moves on with speedy pace.
Exit Knopp happily. Enter Knopp's Successor. He writes and the screen displays:
type Container is record
Length : Natural;
Content: String (Some_Range);
end record;
function "=" (Left, Right: Container) return Boolean is
begin
return Left .Content (1 .. Left .Length) =
Right.Content (1 .. Right.Length);
end "=";
type Super_Container is record
C: Container;
end record;
declare
T: constant String := "Lady Ada";
CON_1, CON_2: Container;
SUP_1, SUP_2: Super_Container;
B_Con, B_Sup: Boolean;
begin
CON_1.Length := T'Length; CON_1.Content (1 .. T'Length) := T;
CON_2.Length := T'Length; CON_2.Content (1 .. T'Length) := T;
SUP_1 := (C => CON_1);
SUP_2 := (C => CON_2);
B_Con := CON_1 = CON_2;
B_Sup := SUP_1 = SUP_2;
end;Lady Ada turns in her grave with those activities' meaningfulness. Chorus sings (a tune from "The Bartered Bride" by Bedrich Smetana):
Oh a lot of sweeties are cajoling pussies,
paws as soft as velvet caress you.
Successor starts a program, screen displays:
CON_1 = CON_2 returns True! (qed) SUP_1 = SUP_2 returns False! (@?!)
Lady Ada rises, turns into a tiger and digs her claws into the Successor, while the chorus sings:
But how terrifying when they start applying
tiger claws on you.
Uproar in the audience (Ada programmers), the Producer (Ada Rapporteur Group) dashes on the stage tearing his hair, groans:
When does the next swan leave?
On the screen, the lake reappears, clouds gather, the room becomes dimly lit as before, lightnings flash across the sky. Robert Dewar's thundering voice from off-screen:
I do not see how you could think otherwise... |
|
While the curtains are falling, you can hear Marcel Reich-Ranicki (a famous German literary critic) holding forth:
A very bad play! The mere idea of it is poor, let alone the realization in the Ada Reference Manual 4.5.2(24) - defective. Either something is equal or it is not! |
 |
Ada 95 has removed some restrictions from the language. Now it is legal to define equality for any type. In Ada 83 this was possible for limited types only (since they haven't a predefined equality), which sometimes led people to make types limited just for this purpose, with the unpleasant consequence of losing assignment - you had to define an "assign" procedure instead.
The new rule is meant to be used on types like Container for which the predefined equality incorrectly also considers unused components (those past the range 1 .. Length). The aforementioned function overrides the predefined equality, remedying this deficiency.
But watch out! The predefined equality is not hidden forever. (Ada 2012 will change this behaviour dramatically.) It for instance re-emerges each time when such a type is used in composite types as a component (i.e. in arrays and records) and objects of those types are compared for equality as a whole. It is however important that the type is not tagged; for tagged types, which did not exist in Ada 83, the overridden equality has really disappeared forever, only the new one will be used at all places.
Therefore, in order to prevent re-emergence of the predefined equality when comparing composite objects whose non-tagged components use a redefined equality, you always have to define the following:
type Super_Container is record C: Container; end record; function "=" (Left, Right: Super_Container) return Boolean is ...;
You also have to pay attention to generic instantiations where this effect is present as well. Here you have to take care of transferring the equality as a formal parameter to the generic unit:
generic type T is private; -- Here, predefined equality for T is implicitly defined; -- hence it re-emerges even when overridden. package Reemergence is ... end Reemergence; generic type T is private; -- This declaration prevents re-emergence: with function "=" (Left, Right: T) return Boolean is <>; package No_Reemergence is ... end No_Reemergence;
Which is the origin of this surprising rule in Ada 95 that equality does not compose except for tagged types? The rule is defined in RM 4.5.2(24); the annotated reference manual AARM 4.5.2(24.a) only provides a short hint to upward compatibility with Ada 83.
(This whole discussion does however not apply to language-defined types, cf. RM 4.5.2(32.1/1).)
The language creators (Jean Ichbiah et al.) of Ada 83 intended two ways to handle such cases like Container. Either with each assignment, also fill the unused components with null-values, so that the predefined equality works correctly, which however for large arrays can become quite inefficient; or make Container limited legalizing the definition of "=", at the same time losing assignment, which is equally unfortunate.
There is however a third, albeit hidden, method in Ada 83 to define equality for any type. You have to resort to a trick attributed to John Goodenough. The trick, allegedly well-known and often used, is to derive from a generic formal limited private type and to define equality for this new type, like so:
-- Ada 83, after John Goodenough (also valid in Ada 95, but unnecessary -
-- cum mortuis in lingua mortua)
generic
type Ancestor is limited private;
with function Equal (Left, Right: Ancestor) return Boolean;
package Define_Equality is
type Descendant is new Ancestor;
function "=" -- here legal
(Left, Right: Descendant) return Boolean;
end Define_Equality;Upon instantiation of Define_Equality, the type Descendant inherits all properties of the actual for Ancestor and additionally has equality defined, i.e. (outside of the package Define_Equality) Descendant is not limited if its parent type is not limited.
function Equal -- "=" in Ada 83 here illegal
(Left, Right: Container) return Boolean is
begin
return Left .Content (1 .. Left .Length) =
Right.Content (1 .. Right.Length);
end Equal;
package Container_Equality is new Define_Equality (Container, Equal);
type New_Container is new Container_Equality.Descendant;This new equality is however valid only for objects of the type Descendant and further derived types like New_Container, but not for components of this type in composite objects. Voilà - Ada 95 having to be upward compatible with Ada 83, it obeys this darn rule.
This paper has appeared in Ada Letters Vol XXV Number 3 (September 2005).
See here the German version.
It has since been decided that (at least for records) the decision to let predefined equality re-emerge was a mistake. So the ARG approved AI05-0123-1 on 13-06-2009, which rules that equality for untagged records composes like for tagged ones.
See RM_2012 4.5.2(9.8/3, 14/3, 14.1/3) and the Annotated RM_2012 4.5.2(14.e/3, 39.i/3).
There are good reasons that equality still does not compose for scalar types nor for arrays. See the AI for detailed discussions.
But note that as of today (18-06-2010), Amendment 2 is not yet an ISO standard.
| Contents | © Copyright 2005 C.K.W. Grein |